A Machine Learning Approach for River Pollution Assessment from Camera Images

Abstract

This study focuses on developing a machine learning system for camera-based river pollution assessment using image analysis and Convolutional Neural Network (CNN) architectures. The objective was to design a predictive model, evaluate different CNN architectures, and utilize camera-captured images to predict river pollution. By integrating colour analysis, reflection analysis, turbidity analysis, and object recognition with CNN-based image analysis, a comprehensive and automated method for evaluating river pollution was developed. The study also evaluated several CNN architectures, including MobileNet, CNN Sequential, and VGG16, to identify the most effective model for river health assessment. The results showed that the VGG16 architecture consistently achieved the highest accuracy, thanks to its deep layers and intricate structure that allowed for precise pattern recognition and analysis. The developed machine learning system, integrated with VGG16 and incorporating multiple image analysis techniques, provides a robust and accurate solution for predicting and assessing the pollution of rivers based on camera images. This comprehensive approach enables proactive measures and timely interventions to protect and improve overall river health.